CEIBS Professor Zhang Feida: How Much Future Has Anthropic's $1.5 Trillion Valuation Pre-Spent?

On October 26, 2026, Anthropic, the developer of Claude, publicly filed registration documents for an IPO (initial public offering) in the United States. Anthropic's IPO process has made the $1.5 trillion valuation a focal point of market attention. The company's financial statements are equally noteworthy: revenue is growing rapidly, a massive book net loss and positive operating cash flow appear at the same time; capital expenditures on fixed assets appear modest, yet R&D and long-term computing investment are astonishing, and external capital continues to flow in.

Following the value investing framework from the previous two columns, Zhang Feida, associate professor of accounting at China Europe International Business School, wants to explore in this article: how much of the productivity Anthropic creates can sustainably translate into shareholder returns? And how much of the future has the $1.5 trillion price already prepaid?

In the two articles on SpaceX and Unitree Robotics, I introduced the Ohlson residual income model. In brief: equity value equals book shareholders' equity plus the present value of future residual income; residual income is net profit minus the return that shareholders' invested capital should earn.

Before analyzing with the four-dimensional framework, a set of numbers can make clear what Anthropic has already proven and what capital is betting on.

First, look at the revenue Anthropic has already achieved. Full-year 2025 revenue was $4.585 billion, and in the first half of 2026 it reached $16.286 billion.

Next, look at how much it has spent on growth. In the first half of 2026, its R&D expenses reached $9.741 billion, compared with $1.850 billion in the same period of 2025 and $6.338 billion for the full year of 2025; computing and infrastructure expenses classified by nature of expense were $8.945 billion, and personnel expenses were $3.152 billion.

R&D is classified by use, while computing and personnel are classified by nature; the two expense groups overlap and cannot be added together. These amounts, more than the phrase "huge R&D investment," illustrate the capital intensity of competition in frontier models.

In addition, the three categories of cash flow show another picture of Anthropic's capital expenditure in the first half of 2026.

Putting these numbers together, the company presents two facts that hold simultaneously: on the one hand, operating cash has begun to turn positive and financing scale has risen sharply, with external investors willing to keep providing funds; on the other hand, the real R&D and computing investment is mainly not in the fixed asset purchase line.

Anthropic is an AI company that is relatively asset-light on its statements but economically very capital-intensive. Willingness to invest and financing capacity can corroborate its growth expectations, but they cannot replace the test of returns on new capital.

Next, we will analyze Anthropic's future value from four dimensions: how high ROE (return on equity) is, how long excess returns can last, how much valuable growth can be achieved, and how much risk these returns face.

Anthropic's current capital structure and accounting treatment make it difficult for traditional ROE to directly reflect true capital returns. Following the logic of DuPont analysis, we need to focus on three questions: whether high gross margin can translate into net profit, whether the asset-light model can improve true capital efficiency, and whether large-scale financing can create excess returns.

Profitability (product strength): how much can high gross margin leave behind?

According to the functional classification in the prospectus notes, cost of revenue in the first half of 2026 was $2.380 billion, corresponding to a gross margin of about 85.4%. But in the same period, R&D expenses reached $9.741 billion, and sales and marketing expenses were $3.808 billion, the latter also including channel partner costs.

Therefore, the high gross margin of 85.4% cannot directly represent ultimately high profitability.

A profitable call does not guarantee that the whole company has enough profit. Frontier models require continuous training and R&D, part of which is opening up new capabilities and part of which may merely be maintaining a competitive position. If more R&D must be invested every year to protect revenue, this spending cannot be treated as a cost that will naturally disappear in the future.

The accounting treatment of R&D expenses also affects the interpretation of profit: Anthropic's model development costs are usually expensed when incurred, thus depressing current accounting profit. But R&D may both form technological accumulation with multi-year value and include ongoing costs necessary to maintain model competitiveness; all R&D expenses cannot simply be regarded as one-off investments that can be added back. Judging normalized profit requires looking at both what R&D has formed and how much still needs to be invested each year in the future.

Operating capability: asset-light on the statements does not mean light capital needs

Cloud computing enables Anthropic to mainly use external infrastructure and not have to include all GPUs and data centers in its own fixed assets. Part of infrastructure investment is borne by suppliers, while Anthropic pays its economic cost through cloud computing fees, prepayments and long-term procurement arrangements.

Few fixed assets do not mean little invested capital is needed to serve customers; mechanically dividing revenue by total assets on the statements may also overstate true asset turnover efficiency.

Long-term procurement and leasing arrangements are not all recognized debt, nor should they be mechanically added to other computing announcements. But locking in resources means that if future demand is lower than expected, part of the spending may become a fixed burden that is difficult to adjust; conversely, if paid tasks grow rapidly, securing capacity in advance may also create cost and delivery advantages.

Moreover, this model is not static. Anthropic explicitly expects in its prospectus that in the future it will build more of its own or leased infrastructure, and that capital expenditure and advance payments will increase significantly.

In the future, it may gradually shift from mainly "buying computing power" to partly "building computing power"; unit costs may fall, but fixed assets, depreciation, idle capacity and chip replacement risks will also rise accordingly.

Capital operation capability: after a successful financing, can the money be used effectively?

Amazon is both an investor in Anthropic and a supplier of computing power and a distribution channel. The cooperation announced by the two parties in April 2026 placed new investment and a commitment to purchase more than $100 billion of AWS services over the next decade in the same arrangement. This can help the company expand, but it also requires investors to distinguish financing inflows, computing power procurement and end-customer revenue.

From the perspective of financing cash flow, external capital has indeed cast a strong vote of confidence. In the first half of 2026, net cash inflow from financing activities was $110.774 billion, of which net proceeds from the issuance of redeemable convertible preferred stock were $110.432 billion; the $65 billion Series H financing announced in May corresponded to a post-money valuation of $965 billion, with investors including several large institutions. Compared with net financing inflow of $19.960 billion for the full year of 2025, the scale of funding support has risen markedly.

However, a large financing amount does not mean external institutions unanimously believe the stock is cheap: financing may serve market capture, strategic binding and preferred stock terms, and it does not guarantee that the company can continue to raise funds at the same valuation in the future.

Whether management can invest funds in truly productive computing power, products and customer relationships, and preserve a reasonable equity share for existing shareholders, will ultimately still depend on the return on new capital, not on the scale of financing itself.

Therefore, in addition to traditional DuPont analysis, I prefer to add an "adjusted return on invested capital" (ROIC) as an auxiliary judgment: looking at both the equipment the company purchases itself and recognized lease right-of-use assets, operating prepayments, and the technological accumulation from R&D that may have multi-year value; and separately conducting a cash payment scenario analysis for future minimum purchase commitments.

Financial conclusion

The key to measuring Anthropic's return on capital is not the low capital expenditure and superficial asset turnover ratio on the statements; rather, it is whether all economic inputs—R&D, leased computing power, self-built infrastructure and long-term commitments—can continue to create returns exceeding the cost of capital①.

II Sustainability: How long can model leadership last?

The Ohlson framework cares about how long excess returns can last, that is, how deep the moat is. For Anthropic, leadership in a certain generation of models is an advantage; whether it can maintain customer relationships and profit margins across multiple rounds of technological iteration is what determines the thickness of the valuation.

There are at least two layers of barriers here: the first layer is model capability, reliability and cost of use; the second layer is the switching cost formed after the model enters the customer's codebase, permission system, business processes and compliance requirements. The deeper the latter layer, the more likely customers are to pay for continued service, but customers' adoption of multiple models and the greater ease of replacing underlying interfaces may also weaken this stickiness.

The benefits of cost reduction also need to be shared between the company and customers. When Anthropic released Opus 5.5 in September, it reduced input and output unit prices by 20% compared with the previous generation, and said customer costs for typical tasks could be reduced by about 40%. This shows that technological progress can expand demand and will also be passed on to customers through price cuts. How much profit the company ultimately retains depends on the relationship between the decline in internal unit costs, the decline in selling prices and the growth in usage.

Financial conclusion

A sustainable moat must withstand the test of model replacement and price declines, and ultimately be reflected in customer retention and return on capital. The financial value of a technological advantage depends on how long it can be maintained.

III Growth: from assisting work to completing tasks

Products such as Claude Code extend the growth space from chat to actual work. If AI moves from giving a piece of advice to further completing a development, analysis or operations task, the amount customers are willing to pay may have the opportunity to be linked to the value created by the task, rather than depending only on the price of a software seat.

This also changes the way growth is tested. But growth in usage does not necessarily equal growth in value. More usage may also come from longer reasoning processes or repeated attempts. What really needs to be observed is the total cost for customers to complete a task, their willingness to pay, and Anthropic's incremental return after deducting computing power and ongoing R&D investment.

If efficiency improvements reduce customer costs and bring more tasks, while making revenue grow faster than the investment needed to maintain competition, there is room for residual income to expand.

Conversely, if new revenue requires a proportional increase in computing power, R&D and delivery support, the return on capital may not improve; whether growth creates value still depends on whether the return on new investment can exceed the cost of capital.

Management's locking in resources in advance shows strong expectations for future task demand; external institutions' willingness to provide large amounts of funding shows that the capital market and strategic partners support this expectation. But both are expectations, not returns: only when new paid tasks, computing power utilization, unit economics and operating cash recovery continue to keep up can it be shown that growth investment is being converted into sustainable value.

Financial conclusion

Whether growth creates value depends on whether all productive inputs—including expensed R&D, leased computing power and self-built infrastructure—can bring repeatable revenue and incremental returns above the cost of capital.

IV Risk assessment: How do management, governance and ESG affect valuation?

Management quality: Can Dario Amodei balance strategic ambition and investment discipline?

Dario previously served as Vice President of Research at OpenAI, founded Anthropic in 2021, and incorporated safety and reliability into the company's mission. Through its responsible scaling policy and long-term benefit trust, the company has written some of its mission requirements into its institutional arrangements; in 2026, Anthropic became involved in a legal dispute with the U.S. Department of Defense over its opposition to the use of its models for mass domestic surveillance and fully autonomous weapons. These choices show that management is willing to bear real costs for some of its principles, but the policy effects and subsequent implementation still need to be observed.

Dario's risk appetite is more worth analyzing in conjunction with real money. In the same interview, he admitted that rapid technological progress does not mean revenue will necessarily materialize in tandem: if, based on the most optimistic scenario, the company purchased in advance computing power costing hundreds of billions of dollars per year starting in 2027, and demand materialized just one year late, the company could go bankrupt.

Therefore, he would rather bear the opportunity cost of some demand going unmet than control the survival risk of overbetting. This looks more like bounded long-term betting in a high-growth industry than simply putting all the chips on one extreme forecast.

But a sense of mission cannot be directly equated with capital discipline. Long-term procurement and leasing commitments mean the company still bears relatively high risk; safety policy adjustments and the founder's participation in share sale arrangements also remind investors that mission statements, personal shareholding behavior and actual capital allocation outcomes should be examined separately.

Overall, I am currently more inclined to see Dario as an operator with a strong sense of mission, a willingness to make long-term high-risk investments, and an understanding of survival constraints. But this is only a provisional judgment.

The real test will come after the IPO: whether he can adjust computing power investment in a timely manner when demand falls short of expectations, and continue to disclose risks transparently under pressure from profitability and share price—that is the key to testing his long-termism.

Corporate governance: What can public shareholders get, and what can they decide?

Under the IPO arrangements, the co-founders' shareholding entity will have 50.1% of the voting power on ordinary shareholder voting matters, and the long-term benefit trust will in principle have the power to appoint and remove a majority of directors. The company also has public benefit corporation status and needs to balance shareholder interests with public interest objectives.

Governance arrangements will also affect earnings per share. The company separately announced a public benefit equity issuance plan tied to listing and subsequent share price milestones.

Therefore, a long-termist governance structure may protect the company's strategic continuity, but public shareholders also need to accept the risks of limited control and possible dilution of future per-share interests.

ESG: External impacts have begun to create real costs

The high-profile generative AI copyright dispute has already had financial consequences: the Bartz v. Anthropic case was settled for $1.5 billion and approved by the court in July 2026. In addition, data rights, model misuse, customer safety and energy use for computing power may all change costs, demand or the pace of commercialization. Good safety governance may also increase customer trust and willingness to continue using the service.

Financial conclusion

Governance and operational risks can both affect expected cash flows and raise the return rate required by shareholders. They should be measured separately during valuation to avoid double-counting the same risk.

02

How much of the future does the $1.5 trillion valuation actually pre-spend?

The four-dimensional analysis answered how Anthropic creates value, while the price test can answer how much investors are paying for that value. Following the approach in the Unitree article, we invert the price and translate it into the operating tasks that must be completed in the future.

Net profit required ten years from now = current equity valuation × (1+12%)^10 ÷ 30

Assume investors require a 12% annualized equity return, the market still gives the company a 30x P/E ratio ten years from now, no dividends are paid during the period, no new financing is raised, and the number of shares remains unchanged. For ease of comparison, also assume a normalized net margin of 25% at that time. This is a simplified scenario for reverse-engineering the price, not a company forecast, nor a complete valuation under the Ohlson model (in reality, the company may continue to raise financing and experience dilution).

Under the $1.5 trillion equity valuation scenario, about $155.3 billion in annual net profit would be needed ten years from now. If only a 20% net margin can be retained, the required annual revenue would rise from about $621.2 billion to about $776.5 billion.

This does not prove that $1.5 trillion is unreasonable; rather, it shows that this price requires the company to achieve both sustained rapid growth and a considerable profit margin over the next decade.

We can find a speed reference for this task: according to Reuters, Anthropic disclosed at an investor conference that third-quarter revenue was about $16.8 billion. Simply multiplying that by 4 gives $67.2 billion (this is only a simple annualization of quarterly revenue, not a full-year revenue forecast, nor an official company run-rate metric). Growing from this pace to $621.2 billion would mean annual revenue growth of about 24.9% over the next decade, while also establishing a 25% normalized net margin.

Both assumptions need to be tested. Still giving a 30x P/E ratio ten years from now means the market still recognizes the company's growth and moat at that time; if only 20x is given then, the required annual net profit rises to about $232.9 billion.

If the P/E ratio remains 30x but the return rate required by shareholders rises to 15%, the required annual net profit would also rise to about $202.3 billion.

The table also temporarily ignores changes in the number of shares caused by subsequent financing, equity incentives and public benefit share issuance, and does not separately forecast future capital expenditures, lease payments and R&D needs brought by self-built computing power, so it is not a complete per-share free cash flow valuation.

New capital may create value or dilute the interests of old shareholders; whether a 12% annualized return can be obtained at the current price must be judged separately based on the actual use of funds and the ability to continue investing.

Thus, whether the $1.5 trillion valuation can hold depends on whether growth, profit margin, the duration of competitive advantage and shareholders' equity share can all be realized together. Proving only that AI demand is large is not enough to prove that this price has a margin of safety.

My judgment on Anthropic is: commercial growth already has a real foundation, but the existing evidence is not yet sufficient to conclude that the $1.5 trillion valuation has an adequate margin of safety.

It is not a pure software company that needs only a small amount of capital in the traditional sense; huge R&D, external computing power procurement and a growing number of long-term infrastructure commitments mean that its growth requires continuous investment of large amounts of economic capital. From revenue to sustainable excess returns, there are still R&D efficiency, computing power utilization, capital allocation and equity dilution to overcome.

Therefore, I will not directly reject it because of huge book net losses, nor will I regard $1.5 trillion as a naturally justified price just because it is growing quickly.

What determines its value is not only model capability and revenue growth, but also whether technological leadership can last, whether capital allocation can remain disciplined, and how much of the new profit can ultimately belong to shareholders.

We have already discussed SpaceX's capital investment and Unitree's product strength. This time, Anthropic gives us a new financial perspective: a company whose statements look asset-light may also be operating a capital-intensive business.

For today's investors, only the portion that ultimately remains on shareholders' books determines how much it is worth.

This article comes from the WeChat public account "China Europe International Business School" (ID: CEIBS6688), author: Zhang Feida, published by 36Kr with authorization.